FinMeter is a library for analyzing poetry in Finnish. It handles typical rhyming such as alliteration, assonance and consonance, Japanese meters and Kalevala meter. It can also be used to hyphenate Finnish and analyse meter. In addition, it can do semantic clustering, metaphor interpretation, concreteness scoring and sentiment analysis
pip install finmeter
If you use the methods relating to semantics, metaphors and sentiment, you will need to run:
python3 -m finmeter.download
Sentiment analysis requires tensorflow (tested on 1.9.0 and numpy 1.16.4).
Finnish words can be divided into syllables like so
import finmeter
finmeter.hyphenate("hattu")
>> hat-tu
finmeter.syllables("hattu")
>> ["hat", "tu"]
finmeter.count_sentence_syllables("kissa juoksi")
>> 4
FinMeter can be used to check whether two words rhyme
import finmeter
finmeter.assonance("ladata", "ravata") #True
finmeter.consonance("kettu", "katti") #True
finmeter.full_rhyme("pallolla", "kallolla") #True
finmeter.alliteration("voi", "vehnä") #True
Meters based on the number of syllables can be assessed by FinMeter
import finmeter
finmeter.list_possible_meters()
>> ['tanka', 'kalevala', 'katauta', 'sedooka', 'bussokusekika', 'haiku', 'chooka']
finmeter.assess_meter(u"kissa juoksee\nkovaa juoksee", "haiku")
>> {'verse_results': [(False, '4/5'), (False, '4/7')], 'poem_length_error': '2/3', 'poem_length_ok': False}
The result is a dictionary cointaining information about the meter for each verse in "verse results" and about the overall length in "poem_length_error". Note: For Kalevala you should use analyze_kalevala instead.
Kalevala meter functionality takes the poetic foot into account and accepts verses of upto 10 syllables providing that certain poetic rules are met. In addition, the method assess other features important in Kalevala
import finmeter
finmeter.analyze_kalevala(u"Vesi vanhin voitehista\nJänö juoksi järveen")
>> [{'base_rule': {'message': '', 'result': True}, 'verse': u'Vesi vanhin voitehista', 'normal_meter': True, 'style': {'alliteration': True, 'viskuri': True}}, {'base_rule': {'message': 'Not enough syllables', 'result': False}, 'verse': u'J\xe4n\xf6 juoksi j\xe4rveen', 'style': {'alliteration': True, 'viskuri': True}}]
The method returns a list of analysis results for each verse. If base_rule is True, it means that the verse follows the Kalevala meter, both in syllables and in foot.
To check if a syllable is short, use the following method
import finmeter
finmeter.is_short_syllable("tu")
>> True
The library has a variety of different functions realted to semantics
from finmeter import semantics
semantics.concreteness("kissa")
>> 4.615
semantics.is_concrete("kissa")
>> True
The former method outputs True if the concreteness of the word is equal or greater than 3. The latter method outputs a concreteness score from 1 to 5. Both of the methods will return None for out of vocabulary words.
from finmeter import semantics
semantics.semantic_clusters(["kissa", "koira", "näätä", "hauki", "vesi", "lemmikki", "puhelin", "tieto|kone", "toimisto"])
>> [['koira', 'lemmikki', 'kissa', 'näätä'], ['vesi', 'hauki'], ['toimisto', 'tieto|kone', 'puhelin']]
semantics.similarity_clusters(["koira", "kissa", "hevonen"], ["talo", "koti", "ovi"])
>> 0.18099508
semantics.cluster_centroid(["koira", "kissa", "hevonen"])
>> [-5.84886856e-02 -1.10119150e-03 -3.40119563e-03......]
The library can be used to cluster words together into semantic clusters and to assess the similarity of two word clusters.
The library provides a somewhat functional sentiment analysis, but I wouldn't hold my breath.
from finmeter import sentiment
sentiment.predict("Olipa kakkainen leffa")
>> -2
sentiment.predict("Kaikki on tosi kivaa")
>> 2
The possible values are -2 for strongly negative, -1 for negative, 1 for positive and 2 for strongly positive.
The library can give interpretations for metaphors. The lower the value, the more likely the interpretation. Example for mies on susi
from finmeter import metaphor
metaphor.interpret("mies", "susi", maximum=10)
>> {'A': [('yksinäinen', 0), ('nuori', 3)], 'Adv': [], 'V': [('raadella', 0), ('tappaa', 1), ('ampua', 2), ('liikkua', 2), ('kaataa', 4)], 'N': [('metsästäjä', 1), ('suu', 3), ('vaate', 4)], 'UNK': []}
maximum is an optional parameter to limit the number of interpretations. If you do not need POS tagging, you can pass pos_tags=False.
It is also possible to assess how metaphorical a tenor and vehicle candidates are in a given sentence (tuli leiskuu kuin tähti taivaalla):
from finmeter import metaphor
metaphor.metaphoricity("tuli", "tähti", ["tuli", "leiskua", "kuin", "tähti", "taivas"])
>> 38.90
The score indicates how metaphorical the two words are given a context. NB the sentence has to be tokenized and lemmatized.
If you use this library, cite the following publication
Mika Hämäläinen and Khalid Alnajjar (2019). Let's FACE it. Finnish Poetry Generation with Aesthetics and Framing. In the Proceedings of The 12th International Conference on Natural Language Generation. pages 290-300